Skip to content

Team Profiles

A Team Profile in Kubeflow creates a dedicated workspace for your team.

Each team gets their own isolated environment with specific resource quotas (CPU, memory, storage, etc.).

Getting a Profile

To request a new team profile, please open a SNOW ticket here.

In your request please include:

  • Your team name
  • Team owner's email
  • Team members's email

    Note: The team admin can also add/remove users after profile creation

  • Resource requirements (mostly for GPUs)
  • The path to your project's EOS home (e.g /eos/project/m/mlops)
  • Brief description of your use case
# Example of resources allocated to a team profile
resourceQuotaSpec:
  hard:
    cpu: "16"
    memory: 64Gi
    requests.nvidia.com/gpu: "4"

Obtaining an authentication token for EOS

EOS authentication is done automatically for individual profiles. In the case of team profiles, you must obtain a token manually, that will be refreshed automatically.

Warning

The script provided below is a temporary workaround. It will soon be replaced by the auth-get-user-token CLI

  • Create and open a notebook in your team profile
    • Disable all the advanced options for the notebook
      • In the notebook creation page, under advanced options untick all of the following:

Advanced Options

  • Ensure python 3 and the requests package are installed (pip install requests)
  • Run this script to generate the token
  • Authenticate with the service account credentials
  • Copy the response and replace TOKEN_FILE_CONTENTS below. Keep in mind, to keep the single quotes. 
    apiVersion: v1
kind: Secret
metadata:
  name: oauth2-session
  annotations:
    oauth2-refresh-controller.cern.ch/is-token: "true"
    oauth2-refresh-controller.cern.ch/audience: eos-service
stringData:
  oauth2: 'TOKEN_FILE_CONTENTS'
  clientID: "kubeflow-ml"
  • In your notebook, apply your secret: kubectl apply -f mysecret.yaml